Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!ukc!harrier.ukc.ac.uk!gos.ukc.ac.uk!dac
From: dac@ukc.ac.uk (David Clear)
Newsgroups: comp.sys.atari.st
Subject: Re: Fine Scrollin'
Message-ID: <2823@gos.ukc.ac.uk>
Date: 20 Feb 90 23:08:26 GMT
Reply-To: dac@ukc.ac.uk (David Clear)
Organization: Computing Lab, University of Kent at Canterbury, UK.
Lines: 416

Recently there has been a query about smooth scrolling on the ST and the
technique of using overscan was mentioned. Well, here is my offering. The
source code below is in 68000 assembler and implements the overscan
technique. The code allows physbase to be positioned on any 8 byte
boundary.

I hope the code is self explanatory. The overscan code takes 20 scan lines
to play with overscan and 2 for overheads. To remove this loss from the normal
playing area I've put the overscan in the top border. This gives rise to
two `features':
	a) The screen is now around 208 lines long.
	b) Screen memory is offset by two bytes.

So, your screens start at address $78002, $7800a, etc.

The 208 lines bit can be sorted out as it's just a matter of when the palette
is re-loaded. Just hack off the end of the code and set a raster interrupt
(timer B) to restore the palette.

The program loads a .pi1 file into screen memory (defined as $78002) and
allows you to scroll around it using the cursor keys. The only way out is
to reset the machine. Sorry, but I use a program I wrote that throws me back
into my assembler (DevpacST) on pressing RESET.

You can play with the code to your heart's content. If you know any games
programmers who don't have scroll code then please make sure they get it.
The new STE has a video address low byte. The old ST doesn't. We NEED
high speed action on STs. This and similar code is a step towards it.

		PROGRAM SOME FAST GAMES NOW!

Imagine Speedball with hardware scrolling...

This code gives smooth vertical scrolling. Horizontal scrolling can be done
by building N screens and flipping between them. For example, 8-pixel
scrolling can be done with 2 screens, 4-pixel scrolling with 4 screens, etc.

Figure it out. There may be alot to keep track of but it'll be DAMNED fast.


One last thing. And it's important.

The top border code runs from timer D. The timing of the interrupt is critical.
There are potential problems:
	1) The interrupt may be delayed whilst a long 68000 instruction is
	   executed. Something like a divide or movem with LOTS of registers.
	   I don't know if this will effect the code but if something goes
	   wrong then careful time planning can solve the problem.
	
	2) Another MFP interrupt may trigger just before the timer D one.
	   This is a SERIOUS problem. The interrupt acknowledge all by itself
	   will throw the timing out. The solution is, fortunately, quite
	   simple(ish):
		Set another interrupt to go off just before the timer D one.
		In it, disable all other MFP interrupts. At the end of the
		scroll code (after re-loading the palette), re-enable the
		interrupts. This should be done using the mask registers
		rather than the enable registers.

	3) Along similar lines: As timer D is set up from the VBI, if the
	   VBI is delayed by an MFP interrupt then, again, the timing will
	   go wrong. A solution works along the same lines as for (2):
		Assuming the scroll code is solid, at the end of it start
		another timer which will interrupt just before the VBI. In
		it, disable the MFP interrupts. After timer D initialisation
		in the VBI, re-enable the interrupts.

	4) The scroll code takes around 1.4ms (about 7% processor time).
	   During this time no keyboard interrupts will be handled. It is
	   possible, therefore, to lose keyboard packets. There is a
	   solution:
		The border code is full of nops. These are for timing. Each
		nop gives an 8 cycle delay. During the longer nop delays
		it is easy to put in code which checks the 'buffer full'
		bit of the keyboard ACIA and stores the data byte if
		appropriate. Be sure to calculate cycle times accurately
		as a single timing miscalculation will produce all sorts of
		weird effects.

Although these problems have to be ironed out the technique is VERY powerful.

Enjoy,

	Dave.

PS. As it is the program will not assemble as a line has been commented
    out. The line is:
		
* .fname	dc.b	'picture.pi1',0

The asterisk should be removed and the quotes filled with the name of your
favourite .PI1 picture.

------------------------- GO FOR IT ------------------------

* This code implements hardware scrolling on an Atari ST (NOT STE).
* This code is (c)1990 David Clear and is free.
* Feel free to use this code or modifications of it for whatever
* you so desire - commercial or private.
*
*	NOW LET'S HAVE SOME DAMN FAST NEW GAMES!
*
* Note:
*	A 'feature' of this code is that it displays around 208
*	scan lines instead of 200. If you want to force 200 scan lines
*	then you can:
*		a) Re-write the top border code and extend the
*		   timer D delay.
*		b) Remove the palette re-load at the end of the
*		   scroll code and set a raster interrupt lower
*		   down to re-load the palette.
*
* Important routines:
*	myvbi	- Timer D startup code & register loading.
*	topb	- Top border scroll code.
*	setscrn	- Screen address setup.

* Macro to create a given number of NOP instructions.
* Used for timing.
* eg:
*	nops	4
* creates:
*	nop
*	nop
*	nop
*	nop
*
* 1 nop is 8 cycles. Long numbers of nops can be replaced by dbra loops
* and the like.
nops	macro
	dcb.w	$4e71,\1
	endm

* Picture data
SCREEN	equ	$78002			* Address of image
PICTURE	equ	SCREEN-34		* Address to load .PI1 picture
PALETTE	equ	PICTURE+2		* Address of palette

* Hardware labels
vbi_vec		equ	$70
timerd_vec	equ	$110

vid_baseh	equ	$ffff8201
vid_basem	equ	$ffff8203
vid_countl	equ	$ffff8209
vid_sync	equ	$ffff820a
vid_shift	equ	$ffff8260
vid_palette	equ	$ffff8240

ikbd_data	equ	$fffffc02

mfp_iera	equ	$fffffa07
mfp_ierb	equ	$fffffa09
mfp_isrb	equ	$fffffa11
mfp_imrb	equ	$fffffa15
mfp_tcdcr	equ	$fffffa1d
mfp_tddr	equ	$fffffa25

* This just demonstrates the routine. It is non-exitable. You have
* to reset to get out of it. Nevermind.
start	lea	vars,a6			* a6 -> variables
	clr.l	-(sp)			* Go into supervisor mode
	move.w	#$20,-(sp)
	trap	#1
	addq.w	#6,sp
	move.l	#mystack,a7		* Set my stack pointer
	clr.w	-(sp)
	move.l	#-1,-(sp)		* Set low res.
	move.l	#-1,-(sp)
	move.w	#5,-(sp)
	trap	#14
	lea	10(sp),sp
	bsr	drawscr			* Draw the screen
	move.w	#$2700,sr		* Kill interrupts
	move.l	#SCREEN,d0		* Set the screen address
	move.l	d0,screen(a6)
	bsr	setscrn
	clr.b	(mfp_iera).w		* Clear IERA
	clr.b	(mfp_ierb).w		* Clear IERB
	andi.b	#$f8,(mfp_tcdcr).w	* Stop timer D
	bset.b	#4,(mfp_ierb).w		* Enable timer D
	bset.b	#4,(mfp_imrb).w		* Mask enable timer D
	move.l	#myvbi,(vbi_vec).w	* Set VBI address
.die	stop	#$2300			* Enable VBI
	bra.s	.die			* Forever and ever...

* Load a .PI1 screen into screen memory
drawscr	clr.w	-(sp)
	pea	.fname(pc)		* Fopen()
	move.w	#$3d,-(sp)
	trap	#1
	addq.w	#8,sp
	move.w	d0,d3
	move.l	#PICTURE,-(sp)
	move.l	#32066,-(sp)
	move.w	d0,-(sp)		* Fread()
	move.w	#$3f,-(sp)
	trap	#1
	lea	12(sp),sp
	move.w	d3,-(sp)		* Fclose()
	move.w	#$3e,-(sp)
	trap	#1
	addq.w	#4,sp
	rts
* Replace this string with a suitable Degas file
* .fname	dc.b	'picture.pi1',0
	even

* This is the VBI code. The VERY FIRST thing to be done is to set up
* the timer D interrupt.
myvbi	move.l	#topb,(timerd_vec).w	* Set timer D interrupt vector
	move.b	#$65,(mfp_tddr).w	* Set timer D data register
	andi.b	#$f8,(mfp_tcdcr).w	* Set timer D subdivider mode
	ori.b	#4,(mfp_tcdcr).w	* Start timer
*---------------------------------------*
* This code is not time critical but should stay here
	movem.l	d0-d7/a0-a6,-(sp)	* Save all the registers
	move.b	bscan(a6),d0		* Move the overscan registers
	ext.w	d0			* into working registers
	move.w	d0,wbscan(a6)
	move.b	rscan(a6),d0
	ext.w	d0
	move.w	d0,wrscan(a6)
	move.b	nscan(a6),d0
	ext.w	d0
	move.w	d0,wnscan(a6)
*---------------------------------------*
* Now just a little keyboard control code.
* VERY poorly written, but what do you want for nothing?
	move.b	(ikbd_data).w,d0	* Get key code
	cmpi.b	#$4b,d0			* Left arrow?
	bne.s	.nl			* No
	subq.l	#8,screen(a6)		* Screen = screen - 8
	bra.s	.yes			* Set the address
.nl	cmpi.b	#$4d,d0			* Right arrow?
	bne.s	.nr			* No
	addq.l	#8,screen(a6)		* Screen = screen + 8
	bra.s	.yes			* Set the address
.nr	cmpi.b	#$48,d0			* Up arrow?
	bne.s	.nu			* No
	subi.l	#160,screen(a6)		* Screen = screen - 160
	bra.s	.yes			* Set the address
.nu	cmpi.b	#$50,d0			* Down arrow?
	bne.s	.no			* No
	addi.l	#160,screen(a6)		* Screen = screen + 160
*---------------------------------------*
* Call setscrn to set the screen address
.yes	move.l	screen(a6),d0		* Set the screen address
	bsr	setscrn
*---------------------------------------*
* What is done finally is to set the video chip's screen pointers.
* This should be done after the call to setscrn.
* NOTE:
*     The physbase register is latched on the VBI. This means that
* this change will not effect the screen about to be drawn but will
* effect the following screen. Anyone who uses double buffered screens
* will know this already.
.no	move.b	vidh(a6),(vid_baseh).w	* Set the video pointer
	move.b	vidm(a6),(vid_basem).w
*---------------------------------------*
	movem.l	(sp)+,d0-d7/a0-a6
	rte

* This function sets the screen base address. Well, it doesn't
* really load the video pointers, it just sets up variables which
* the vbi will use.
*---------------------------------------*
* IMPORTANT
* ---------
* The screen can be moved on 8 byte boundaries. There is a 'feature'
* of the top border code that means the screen is shifted two bytes.
* This means that if you call this function with #$78000 in d0, the
* REAL beginning of your screen will be $78002.
*---------------------------------------*
* Set screen address
* Input:
*	d0.l = Address to put screen.
setscrn	subi.l	#320,d0			* The magic offset
	move.b	d0,d1			* Save low byte
	lsr.l	#8,d0
	move.b	d0,vidm(a6)		* Store mid byte
	lsr.l	#8,d0
	move.b	d0,vidh(a6)		* Store high byte
	lea	normal(pc),a0
	lsr.b	#3,d1			* Multiples of 8
	ext.w	d1
	move.b	0(a0,d1.w),nscan(a6)	* Normal scan lines
	move.b	right-normal(a0,d1.w),rscan(a6) * Right overscan
	move.b	both-normal(a0,d1.w),bscan(a6)  * R & L overscan
	move.b	offset-normal(a0,d1.w),d0	* Video base offset
	sub.b	d0,vidm(a6)		* Subtract from video base
	bcc.s	.ok
	subq.b	#1,vidh(a6)
.ok	rts

screen	rs.l	1

* Variables used by the code
vidh	rs.b	1		* Video address
vidm	rs.b	1
rscan	rs.b	1		* Number of lines of overscan
bscan	rs.b	1
nscan	rs.b	1
wrscan	rs.w	1		* Working variables
wbscan	rs.w	1
wnscan	rs.w	1

* This is the scroll code. It uses overscan to allign the screen.
* The overscan takes 20 scan lines but 22 are required to make things
* look nice. So that the 22 scan lines aren't eaten from the play area
* the top border is removed and the overscan done in there.
* IMPORTANT:
* The top border removal shifts the screen by 2 bytes.
zeroes	dc.l	0,0,0,0,0,0,0,0		* Zeroes for palette
topb	andi.b	#$f8,(mfp_tcdcr).w	* Stop timer D
	movem.l	d0-d7/a0-a3,-(sp)	* Save some registers
	movem.l	zeroes(pc),d0-d7	* Clear the palette
	movem.l	d0-d7,(vid_palette).w
*---------------------------------------*
* Load some important registers
	lea	(vid_sync).w,a0		* a0 -> video sync
	lea	(vid_shift).w,a1	* a1 -> video mode
	lea	(vid_countl).w,a2	* a2 -> video low byte counter
	moveq	#0,d0			* 60Hz/Low res
	moveq	#2,d1			* 50Hz/High res
*---------------------------------------*
* Pick up the number of respective overscan lines needed
	move.w	wbscan(a6),d3		* Left & right overscan
	move.w	wrscan(a6),d4		* Right overscan
	move.w	wnscan(a6),d5		* No overscan
*---------------------------------------*
* Remove top border
	lea	.skip(pc),a3		* a3 -> Sync buffer
	move.b	d0,(a0)			* 60Hz
	nops	81			* Wait for next line
	move.b	d1,(a0)			* 50Hz
*---------------------------------------*
* Sync on raster
.loop	tst.b	(a2)			* Screen running?
	beq.s	.loop			* No
 	move.b	(a2),d2			* Get low byte
	adda.l	d2,a3			* Add to sync buffer
	jmp	(a3)			* Jump into sync buffer
.skip	nops	100			* Sync buffer
*---------------------------------------*
* This code performs left and right overscan
	bra	btest			* Branch to overscan test
lboth	nop
	move.b	d1,(a1)			* You really think I can
	move.b	d0,(a1)			* be bothered to comment
	nops	89			* all these nops and
	move.b	d0,(a0)			* loads? You must be out
	move.b	d1,(a0)			* of your mind.
	nops	13
	move.b	d1,(a1)
	nop
	move.b	d0,(a1)
	nops	9
btest	dbra	d3,lboth			* Do next line
*---------------------------------------*
* This code performs right overscan
	bra	rtest			* Branch to overscan test
lright	nops	87
	move.b	d0,(a0)
	move.b	d1,(a0)
	nops	13
	move.b	d1,(a1)
	nop
	move.b	d0,(a1)
	nops	16
rtest	dbra	d4,lright
*---------------------------------------*
* This code does nothing but is used for timing the palette re-load.
* If you only want 200 raster lines then just remove this code and
* set up a raster interrupt for a few lines down.
* DON'T remove the d5 loading code at the beginning as this may throw out
* the top border removal code. If it does then increment the number of
* nops in the top border code.
	bra	ntest
lnorm	nops	125
ntest	dbra	d5,lnorm	
*---------------------------------------*
* Reload the palette
	nops	75			* Wait for end of line
	movem.l	PALETTE,d0-d7
	movem.l	d0-d7,(vid_palette).w
end	movem.l	(sp)+,d0-d7/a0-a3	* Restore registers
	bclr.b	#4,(mfp_isrb).w		* Clear MFP ISR bit
	rte

* This is the magic table which makes it all possible.
* DON'T MESS WITH IT!
normal	dc.b	9,10,4,5,6,0,16,10,11,12,13,0,1,2,3,4
	dc.b	20,14,0,1,10,11,12,6,0,1,2,3,12,13,0,0
right	dc.b	5,2,8,5,2,8,4,10,7,4,1,16,13,10,7,4,0
	dc.b	6,4,1,6,3,0,6,12,9,6,3,8,5,20,0
both	dc.b	6,8,8,10,12,12,0,0,2,4,6,4,6,8,10,12
	dc.b	0,0,16,18,4,6,8,8,8,10,12,14,0,2,0,20
offset	dc.b	$f,$f,$10,$10,$10,$11,$d,$e
	dc.b	$e,$e,$e,$10,$10,$10,$10,$10
	dc.b	$c,$d,$11,$11,$e,$e,$e,$f
	dc.b	$10,$10,$10,$10,$d,$d,$f,$11
	
	even
vars	ds.b	__RS

	even
	ds.l	100
mystack


--------------------------- END ----------------------------
-- 
% cc life.c                      | David Clear <dac@ukc.ac.uk>
% a.out                          | Computer Science, University of Kent,
Segmentation fault (core dumped) | Canterbury, England.